The title should probably be “BCM4908 vs IPQ8072A, which has a higher packet forwarding rate?” The models tested were Xiaomi AX9000, Asus AX86U, TP-LINK XDR6060, all of which have 2.5G WAN ports and no 1Gbps bottleneck in total bandwidth to all LAN ports. This is to test whether the small packet forwarding rate of the 2.5G network port can be full line speed.
Packet forwarding rate refers to how many 64byte packets can be forwarded in one second, the unit is pps. If it is a gigabit rate network port, the full line speed is 1.488Mpps, if the 2.5G network port, multiplied by 2.5 times is, is equal to about 3.72Mpps.
The software platform for testing the packet forwarding rate is minismb, a republic of three computers, a console, two servers, two servers are equipped with X550-T2 network card (with a port) and I350-T4 network card (with three ports)
The connection schematic of each device is as follows: (ASUS AX86U)
These models have only one 2.5G network port, so the other server needs to connect to the router’s three LAN ports with multiple network ports, or it won’t meet the bandwidth of the 2.5G port. In short: 2.5G WAN to (LAN1+LAN2+LAN3). The three LAN ports receive packets from the WAN port at the same time, you need to use the “port forwarding” function, set the router’s WAN port parameters and port mapping can play through the data, here is not a detailed explanation of the steps.
Port forwarding is set up as follows.
ASUS AX86U test results: (2.5G WAN downlink, 64byte small packet)
Look at the sum of the RX data of the back three ports (line 5): 1.458Mpps. There is a little difference with the theoretical value of 1.488Mpps, because my test platform can only run to this value, so 1.458Mps can be said to reach the line speed of the gigabit port. But far from the 2.5G line speed.
Now look at the reverse: (2.5G WAN uplink, 64byte packet)
Also look at the data in line 5 RX, three LAN ports send 64byte packets to the 2.5G WAN port at the same time, the total conversion rate is also the line speed of the gigabit port. At this point there is a core CPU occupation has been full, can not say that it did not make all the effort, right.
If you send packets to only 2 LAN ports, the total sum is only about 1.45Mpps. As shown in the figure below.
Strange! How come both directions are not up to par? Is it a bandwidth issue between WAN and LAN ports?
No, when using large packets (1518byte), it can nevertheless reach full line speed. It can also reach full speed when forwarding in both directions at the same time, as shown below: (1518byte, both directions)
The total speed is RX Kbps in line 7, the sum of the three LAN ports is 2.4Gbps, and the WAN uplink speed is 2.5Gbps. here it proves that there is no bottleneck in the bandwidth, and the transmission speed can reach 2.5Gbps.
There are two NAT types in ASUS routing are Symmetric NAT, Full Cone NAT, set up and retested respectively, the results did not change.
The same BCM4908 + 2.5G network port XDR6060, test results are very consistent with the ASUS AX86U: (64byte, WAN port downstream)
Both directions are around 1.45Mpps. That is, it only achieves the small packet forwarding performance of a gigabit port!
BCM4908 is currently the most powerful CPU, more than this little forwarding performance, right? Asus AX11000, Netgear RAX200 is also the same? I don’t know, I don’t have one!
Now test Xiaomi AX9000 (CPU is IPQ8072A) to see the small packet forwarding rate: (64byte, WAN port downstream)
The sum of data from the three network ports behind the RX line is 3.562Mpps, reaching the limit of my tester platform (2.5G port line speed is 3.72Mpps).
64byte packets in both directions at the same time to see.
Looking at the RX data in line 5, it reaches 3.574Mpps in both directions.
The IPQ8072A’s 64byte packet forwarding rate also reaches full wire speed on the 2.5G network port, as well as in both directions.
To briefly summarize the packet forwarding rates of the three routers.
In fact, do not pay too much attention to this small packet forwarding rate, daily use of indifferent, unless the forwarding rate is very low very low.
BCM4908, I do not believe that it only this forwarding performance, unless it really only this performance. The only possibility I can think of is money. The vendor gives Broadcom some more money, Broadcom gives more performance.
Is there a problem with the test style party? The same way and steps set on the IPQ8072A but no problem ah! And repeatedly tossed for three days. Not just a model, there is another XDR6060 is also exactly the same result!